Why do Harman and Sonarworks sound so different?

[Volutrik]

Why do these two target curves sound so different from each other? I know it's about the methodology - Sonarworks uses a different dummy head to perform the measurements. They also include how the body perceives bass impact in the calibration, that's why it sounds do different.

But wait. Isn't the goal of both to sound flat? I mean, even if they're using different methodologies, if the result can't be the same, it should at least be close, right?
In my overall experience comparing the two, I find Harman way brighter than Sonarworks.
"So maybe I just have to compare them again to know if what I'm hearing is really true" - then Sonarworks sounds like it has too much 2-5kHz when compared to Harman.
"Ok, maybe one more comparison" - then Harman sounds like piercing the ears/hollow.
Because of this never ending comparison, without a reference of how flat really sounds like, I just don't know what target to trust. I've never been into a mixing room nor have listened to calibrated speakers.

The quest can't stop, right?
I've been thinking about getting Genelec's expensive Aural ID, to see if I can come close to hearing a flat sound, but then the issue would still be there: the headphone's damn coloration.
Even worse! To my understanding, because the HRTF produces a particular frequency response, using a pre-defined calibration like Harman or Sonarworks is probably going to do more harm than good, because they're using a different head/ears than mine to perform the measurements and the frequency response would disalign the one introduced by my HRTF.

A light at the end of the tunnel. A really far one, but hey, it's there. I think my search for a flat sound is over until I get some flat-measuring speakers and acoustically treat my room for this purpose.

But wait, this can't be over! Is this really the end of the search for hearing a flat sound with headphones? How could one know which curve to trust if they have no real reference of a flat sound? Or one that's closest to flat, at least?

 

[solderdude]

As you already answered yourself they use different measurement methods and have different targets because they both have done their own research.
Both methods may not lead to satisfactory results.

How 'flat' something sounds also depends on the recording and listening level.
How a 'flat' speaker sounds in your room may differ from other rooms.
How well a headphone couples to your ears may well be different from measurement standard fixtures.

Look for speakers and headphones you enjoy. In the end that's what you will prefer and use most.

 

[holdingpants01]

There's a great white paper about it on their website, highly recommended: https://www.sonarworks.com/blog/research/white-paper

 

[solderdude]

The paper has this conclusion:

We argue that our data shows that there is no one sound to be liked by everyone. It is not that the industry does not know enough about consumer preferences, but rather the fact that consumers do not converge on one target.

I agree with that. Harman target is valid (liked by) the majority of listeners and thus makes sense to try to tune headphones to that tonal balance when you want to sell the most headphones and get the most positive reviews/feedback.

In practice there are many folks that prefer a divergent target from that Harman target. The delta is shown below and is quite substantial.

So there is no one single correct headphone measurement fixture nor is there a single correct 'target' but there is most preferred tonality.
As there is a difference in test fixtures as well and headphones, after correction with the exclusion of tonal preference target, also differs between the same headphone on different test fixtures it is very clear that headphone measurements are at best 'indicative' and that some fixtures are slightly more suited (have better convergence with perceived tonality for the majority of users) than other fixtures.

In any case... there is no real standard for headphone measurements and no single correct 'target' either. Just research showing that the majority (but certainly not all) listeners have a somewhat similar preference in tone.

One can use a target as 'starting point' and change that to one's personal preference. There is nothing wrong with that.

 

[wwenze]

One guy prefers his violinist in a room, the other prefers his in a concert hall. Same violin, same guy btw, just to clear out any misunderstanding. Also playing the same genre of music. Maybe one had coffee and the other had beer, I'm not sure.

 

[somebodyelse]

Headphones interact with the individual wearing them. If you measure the individual you can probably come up with a 'correct' target for them. If you can't measure the individual the best you can do is have a target that matches most of the population. For most people it probably won't be too far off but there will be outliers.

https://www.audiosciencereview.com/...presentation-on-headphone-measurements.31997/

 

[muslhead]

The paper has this conclusion:


I agree with that. Harman target is valid (liked by) the majority of listeners and thus makes sense to try to tune headphones to that tonal balance when you want to sell the most headphones and get the most positive reviews/feedback.

In practice there are many folks that prefer a divergent target from that Harman target. The delta is shown below and is quite substantial.

So there is no one single correct headphone measurement fixture nor is there a single correct 'target' but there is most preferred tonality.
As there is a difference in test fixtures as well and headphones, after correction with the exclusion of tonal preference target, also differs between the same headphone on different test fixtures it is very clear that headphone measurements are at best 'indicative' and that some fixtures are slightly more suited (have better convergence with perceived tonality for the majority of users) than other fixtures.

In any case... there is no real standard for headphone measurements and no single correct 'target' either. Just research showing that the majority (but certainly not all) listeners have a somewhat similar preference in tone.

One can use a target as 'starting point' and change that to one's personal preference. There is nothing wrong with that.

Can you describe this graph you included because without any annotation i cant figure out its purpose in your post.
Or, if you want to just link a page where the details are provided, that works too.
Thanks

 

[AdamG]

The Harman Preference Curve is an average preference of surveyed participants. So since it is just an average preference there will always be folks who want more Bass, or less base or more treble and so on. It’s just an Average from a sample group. It’s a “Preference” curve, not a target. Final fine tuning with EQ and low distortion headphones are what counts. We all have different hearing capabilities and different listening tastes. Don’t get all wrapped around the axle of a “Preference” curve.

Look for headphones that have low distortion and experiment with EQ/PEQ and salt and pepper to taste. ‘Low distortion across the audible range 20hz to 16khz (give or take based on your personal hearing range). If you have a headphone with low distortion measurements then you can EQ to taste without worrying about driving the speakers into distortion.

These Preferences Curves are really focused towards the Manufacturers so they know what the average majority may want the headphones tuned towards out of the box. It’s really meaningless to consumers as we will all normally have our own tastes/preference on actual tuning and that’s why we are always recommending utilizing EQ/PEQ to achieve the best possible sound that suits you.

The above is just my humble opinion.

 

[holdingpants01]

Can you describe this graph you included because without any annotation i cant figure out its purpose in your post.
Or, if you want to just link a page where the details are provided, that works too.
Thanks

It's from the paper which I posted right before

 

[solderdude]

Can you describe this graph you included because without any annotation i cant figure out its purpose in your post.
Or, if you want to just link a page where the details are provided, that works too.
Thanks

See post #3

When you want to download that Sonarworks info in pdf format you can download that here

It shows the delta for preference.

 

[SSS]

There's a great white paper about it on their website, highly recommended: https://www.sonarworks.com/blog/research/white-paper

The white-paper is not really helpful. Heads, ears are so different between people that there will probably be no common standard for measurement. And even if measured similar, the sound will still be different like with loudspeakers. An individual should try several models from diferent brands which is best for the kind of music listening.

 

[solderdude]

That's what the paper tries to explain.

 

[SSS]

That's what the paper tries to explain.

Sure, I did not read all the text. After doing, there is a lot of evaluation results with the outcome to standardize for a certain f-curve which is somekind of an average. But this is perhaps helping earphone designers and producers. For a customer I see little help from this paper. As an oldie I anyway do not understand why listening music with earbuds.

 

[solderdude]

Earbuds make little sense to me.
Over-ear, on-ear and in-ear is a totally different matter though.
Any of these can be quite high quality.

The paper is not for your average headphone consumer though, they would not understand one word in it.
It may offer some clarity (pun intended) as to why the Harman and Sonarworks targets differ.

 

[SSS]

Earbuds make little sense to me.
Over-ear, on-ear and in-ear is a totally different matter though.
Any of these can be quite high quality.

The paper is not for your average headphone consumer though, they would not understand one word in it.
It may offer some clarity (pun intended) as to why the Harman and Sonarworks targets differ.

Agree. Not to forget through bone listening. Pesonally I listen not often with my Stax electrostats since earphones are not naturally to me. Best usage is for me to analyze the details qualtity in music regarding resolution, instruments spacing and possible distortions and/or noise.

 

[Volutrik]

As you already answered yourself they use different measurement methods and have different targets because they both have done their own research.
Both methods may not lead to satisfactory results.

How 'flat' something sounds also depends on the recording and listening level.
How a 'flat' speaker sounds in your room may differ from other rooms.
How well a headphone couples to your ears may well be different from measurement standard fixtures.

Look for speakers and headphones you enjoy. In the end that's what you will prefer and use most.

Hey! I actually thought about something that can potentially be the final answear for getting the true flat sound:

1- You flatten out the frequency response of a headphone measurement done with a GRAS system (because it's the one system compatible with Harman and Diffuse Field target curves). You can easily do that with autoeq.app, there's a flat target available, basically 0db from 20Hz to 20kHz.
2- Now, without any EQ applied, you flatten out the same headphones, this time using your own ears, making sine waves at dfferent frequencies accross the spectrum sound equal in volume (you're basically doing the same thing autoeq.app did in step 1, except that this time you're making it sound flat according to your own hearing).
3- You get the difference (delta) from the GRAS correction and your own self-made custom correction. By doing so, you basically have the last pieces that were left in the puzzle that kept you from having the Harman/Diffuse Field target curve perfectly fit your hearing.
Now you can use the GRAS equalizations available out there + the difference you got from that method to get perfect corrections. You can even pick the delta and create your own Harman/Diffuse Field target. That will allow you to do corrections quicker, because you won't have to do the sine equal loudness procedure everytime you want to calibrate different headphones, you'll just have to specify the target and that's done. In fact, now you can even make your headphones properly sound like others. You can even get your own HRTF if you do that with the Diffuse Field target (or at least get close because of the variance between the fixture's head shape/nose shape). The possibilities are endless.

Do you have ideas on why this wouldn't work? I ask this because I don't have that much knowledge in the field and maybe you or someone else has an answear to why this wouldn't be valid. To me, this seems like the definitive answear, at least on paper

 

[solderdude]

Hey! I actually thought about something that can potentially be the final answear for getting the true flat sound:

1- You flatten out the frequency response of a headphone measurement done with a GRAS system (because it's the one system compatible with Harman and Diffuse Field target curves). You can easily do that with autoeq.app, there's a flat target available, basically 0db from 20Hz to 20kHz.

This already has some issues:
1: You do not own the exact same headphone so production spread is not accounted for and results in errors.
2: You measure against a standard, meaning it is built exactly to certain specifications (build description with tolerances).
3: The Harman curve is merely a guideline and not a standard. In fact there is NO standard for headphone measurements. DF is not the correct target either.
4: The Harman standard is developed with a modified pinna and not the one used on a standard GRAS
5: The GRAS is not really accurate above 6kHz and useless above 8kHz yet most of the issues with sharpness in sound resides above 8kHz.
6: The Harman target is way too 'smooth' above 3kHz so will give measurement errors with a 100% certainty
7: Fit on the head differs from fit on a flatbed, positioning, pad condition, averaging (moving the headphone around and taking multiple measurements), smoothing of the plots.

So... the basis to start from is already suspect to say the least. What the GRAS does is measure according to a defined standard which tells the manufacturer what to build and what the error margin is. Yes, it is based on research of human auditory system but isn't an 'exact' copy of an 'average human ear+ear canal' it is an approximation of it from what the build specs are clearly defined. In that aspect the BK5128 is much more human alike but to really use it for preference measurements (Harman research) it needs to be scrutinized a bit more.

2- Now, without any EQ applied, you flatten out the same headphones, this time using your own ears, making sine waves at dfferent frequencies accross the spectrum sound equal in volume (you're basically doing the same thing autoeq.app did in step 1, except that it's making it sound flat according to your own hearing).

This is extremely difficult to do and requires special test signals. Griesinger has done something like this.

Griesinger's Headphone EQ for Frontal Localization

Has anyone tried this? http://www.dgsonicfocus.com/ It sort of worked for me on the AKG K371. With any headphone or IEM I have a persistent sense of elevation (vertical soundstage) as the frequency goes up. The best I could get with the DG method was a narrower range of elevation. That said...

www.audiosciencereview.com

That method is in frequency 'steps' and not the same as auto-EQ which basically takes a plot and creates the inverse EQ. This is as 'accurate' as the measurement. see the points I made above. That method thus also puts in measurement errors in the EQ.
So.... an easy method... yes.
Sure one can use white or pink noise and faff around with an equalizer but to do this right you really must have a reference when it comes to tonal balance.
Besides..

8: Your pinna and ear canal are likely to be different from the pinna and metal tube.
9: The frequency response of your actual hearing may not be the same as that of the microphone in the coupler.
10: Listening SPL is a substantial factor in tonal balance from a measurement.
11: Not all recordings are made using the same 'target' so music can still sound incorrect.

3- You get the difference (delta) from the GRAS correction and your own self-made custom correction. By doing so, you basically have the last pieces that were left in the puzzle that kept you from having the Harman/Diffuse Field target curve perfectly fit your hearing. Now you can use the normal GRAS equalizations available out there + the difference you got from that method to get perfect corrections. You can even pick the delta and create your own Harman target. That will allow you to do corrections quicker, because you won't have to do the sine equal loudness procedure everytime you want to calibrate different headphones, you'll just have to specify the target and that's done. In fact, now you can even make your headphones properly sound like others. The possibilities are endless.

Basically you use a 'standard fixture' (with all its inaccuracies) combined with a very smoothed target created with a different pinna (not very different but yes, not according to the standard) and God knows how that plot was made (positioning, averaging, conditions) which gives incorrect results above 6kHz and combine that with say the results found using Griesinger (which is, fortunately, not fine grained like the measurement) and expect that delta to have any accuracy.
That delta will differ from headphone to headphone anyway so would have to be done for every headphone.

There is merit to standardization, there is merit to measuring, there is merit to Griesingers method. Not so much sweeping sinewaves and trusting the hearing.
Headphone measurements are not accurate. NONE are.
What they could show you is, when looking at multiple measurements done on different fixtures that should all have the same standard target, maybe DF eventhough that is the wrong one to use is what 'average' deviation exist from 'flat', apply some EQ that undoes this and tune from there to 'taste' (bass and treble levels) so that the result sounds 'realistic'.

Do you have ideas on why this wouldn't work?

Yep, common sense was used above not hindered by the idea that a standard is the truth. It is just a standard. Standards are handy for scientists that like to replicate research.

I ask this because I don't have that much knowledge in the field and maybe you or someone else has an answear to why this wouldn't be valid. To me, this seems like the definitive answear, at least on paper

So no 'sharp' and 'detailed' elaborate EQ requiring many filters and thinking you done a great job as those sharp peaks and dips are very likely very different on your copy on your head.

Why such EQ still 'works':

• The gross deviations are still improved upon despite the errors due to the 'elaborate' EQ so it is still better than no EQ in the 20Hz-6kHz region.
• People believe the available EQ must be the correct one so what is heard MUST be close to the ideal.
• The hearing is quite forgiving.

 

[Volutrik]

This already has some issues:
1: You do not own the exact same headphone so production spread is not accounted for and results in errors.
2: You measure against a standard, meaning it is built exactly to certain specifications (build description with tolerances).
3: The Harman curve is merely a guideline and not a standard. In fact there is NO standard for headphone measurements. DF is not the correct target either.
4: The Harman standard is developed with a modified pinna and not the one used on a standard GRAS
5: The GRAS is not really accurate above 6kHz and useless above 8kHz yet most of the issues with sharpness in sound resides above 8kHz.
6: The Harman target is way too 'smooth' above 3kHz so will give measurement errors with a 100% certainty
7: Fit on the head differs from fit on a flatbed, positioning, pad condition, averaging (moving the headphone around and taking multiple measurements), smoothing of the plots.

So... the basis to start from is already suspect to say the least. What the GRAS does is measure according to a defined standard which tells the manufacturer what to build and what the error margin is. Yes, it is based on research of human auditory system but isn't an 'exact' copy of an 'average human ear+ear canal' it is an approximation of it from what the build specs are clearly defined. In that aspect the BK5128 is much more human alike but to really use it for preference measurements (Harman research) it needs to be scrutinized a bit more.



This is extremely difficult to do and requires special test signals. Griesinger has done something like this.

Griesinger's Headphone EQ for Frontal Localization

Has anyone tried this? http://www.dgsonicfocus.com/ It sort of worked for me on the AKG K371. With any headphone or IEM I have a persistent sense of elevation (vertical soundstage) as the frequency goes up. The best I could get with the DG method was a narrower range of elevation. That said...

www.audiosciencereview.com

That method is in frequency 'steps' and not the same as auto-EQ which basically takes a plot and creates the inverse EQ. This is as 'accurate' as the measurement. see the points I made above. That method thus also puts in measurement errors in the EQ.
So.... an easy method... yes.
Sure one can use white or pink noise and faff around with an equalizer but to do this right you really must have a reference when it comes to tonal balance.



Basically you use a 'standard fixture' (with all its inaccuracies) combined with a very smoothed target created with a different pinna (not very different but yes, not according to the standard) and God knows how that plot was made (positioning, averaging, conditions) which gives incorrect results above 6kHz and combine that with say the results found using Griesinger (which is, fortunately, not fine grained like the measurement) and expect that delta to have any accuracy.
That delta will differ from headphone to headphone anyway so would have to be done for every headphone.

There is merit to standardization, there is merit to measuring, there is merit to Griesingers method. Not so much sweeping sinewaves and trusting the hearing.
Headphone measurements are not accurate. NONE are.
What they could show you is, when looking at multiple measurements done on different fixtures that should all have the same standard target, maybe DF eventhough that is the wrong one to use is what 'average' deviation exist from 'flat', apply some EQ that undoes this and tune from there to 'taste' (bass and treble levels) so that the result sounds 'realistic'.



Yep, common sense was used above not hindered by the idea that a standard is the truth. It is just a standard. Standards are handy for scientists that like to replicate research.



So no 'sharp' and 'detailed' elaborate EQ requiring many filters and thinking you done a great job as those sharp peaks and dips are very likely very different on your copy on your head.

Why such EQ still 'works':

1: The gross deviations are still improved upon despite the errors due to the 'elaborate' EQ so it is still better than no EQ in the 20Hz-6kHz region.
2: People believe the available EQ must be the correct one so what is heard MUST be close to the ideal.
3: The hearing is quite forgiving.

That's exactly what I wanted. Thank you for crushing my expectations with beautiful scientific facts hahahah I think I'll still do a test with this though, just to see if it sounds any better than my tweaked Harman equalization (which I managed to make it sound pretty convincing using Dear VR's Mix plugin, emulating a studio environment)

 

[solderdude]

When you have (calibrated/measured) monitors in a studio at least you have a reference.
This is something most people do not have.
At best they have speakers + room correction (which is not really the same but better than nothing) or just assume the 'flat' speakers they bought are accurate in their room and see that as their reference.

So I would say experiment with the Griesinger method which relies on sounds coming from directly in front of you which differs from 2 speakers with an offset from the middle.
Or you could rely on your gear and try to get the same tonal balance between the headphone and your setup.

You will end up with an EQ that differs from autoEQ but most likely (when headphones are really wonky) will see at least some resemblance.

I suspect the EQ you end up with using your monitors will be much better in your case than the one created by a computer.
The EQ you end up with may not work for others for various reasons.

Experiment the hell out of it and look forward to seeing the results.
Maybe it will even be closer to Sonarworks than Harman because of your application.

 

[Volutrik]

When you have (calibrated/measured) monitors in a studio at least you have a reference.
This is something most people do not have.
At best they have speakers + room correction (which is not really the same but better than nothing) or just assume the 'flat' speakers they bought are accurate in their room and see that as their reference.

So I would say experiment with the Griesinger method which relies on sounds coming from directly in front of you which differs from 2 speakers with an offset from the middle.
Or you could rely on your gear and try to get the same tonal balance between the headphone and your setup.

You will end up with an EQ that differs from autoEQ but most likely (when headphones are really wonky) will see at least some resemblance.

I suspect the EQ you end up with using your monitors will be much better in your case than the one created by a computer.
The EQ you end up with may not work for others for various reasons.

Experiment the hell out of it and look forward to seeing the results.
Maybe it will even be closer to Sonarworks than Harman because of your application.

For sure! I'll keep expetimenting. This subject really interests me, I have fun doing it, even though sometimes I feel a little frustrated hahah It's part of the process anyways and there's progress being made.
I'm thinking of getting an Umik-1 from Cross Spectrum and use it to flatten out my ****** little speakers and then perform the DG Sonic Focus test, so I can achive the best results as possible. It seems like this would be the best solution for now!
Do you know if it's worth getting the Umik-1 from Cross Spectrum or if the one directly from MiniDSP is enough?